1. Forum
  2. Usenet
  3. REC.GAMES.COREWAR

  • He Scans Alone

    From Stephen Gunnell@21:1/5 to All on Sat Oct 29 16:16:47 2022
    Many years ago I was investigating the components of Recycled Bits and I pulled apart the scanner and rebuilt it to be one line shorter. The result was a JMZ scanner with a decoy generator that I called Keshin. Because an ADD.F instruction will change
    both the A and B fields I tried to develop a dual pointer JMN/JMZ scanner. Aoshi was never quite good enough to make the hills but not so bad that I purged the code. Around the same time I was familiar with Mini HSA from Anton Marsden's Electric Head
    which was a fairly unprepossessing JMZ scanner so I never investigated Paul Kline's He Scans Alone (which I guess was another fail :). Fast forward to August 2022. I had just developed some code to approximate the contents of the hill from the warriors
    on hand. This is a really old Idea but it needs modern processor speeds to be effective. And effective it is. One of the old warriors it highlighted was He Scans Alone. When I looked at it there was the Aoshi core but with a brilliant SNE/SUB front end
    that vastly improved the efficiency. So I tinkered ...

    ;redcode-94nop
    ;name He Scans Alone qhoz
    ;author P.Kline
    ;strategy 80% f-scanner switches from SPL to DAT carpet
    ;strategy Tweaked and upcycled by Steve G
    ;assert CORESIZE == 8000

    STEP equ 10
    TIME equ 19
    OFFSET equ 2813

    tPtr dat 100 ,OFFSET ; widely-spaced pointers
    dat 0 ,0
    for 3
    dat 0 ,0
    rof

    tWipe mov tSpl ,<tPtr ; positive wipe of opponent
    mov >tPtr ,>tPtr
    jmn.f tWipe ,>tPtr

    tScan sub.x #0-STEP ,tPtr ; increment and look
    sne *tPtr ,@tPtr
    sub.x *pScan ,@tScan ; increment and look
    jmn.f tSelf ,@tPtr
    jmz.f tScan ,*tPtr
    pScan mov.x @tScan ,@tScan ; swap pointers for attack
    tSelf slt.b @tScan ,#tEnd+4-tPtr ; self-check
    djn tWipe ,@tScan ; go to attack
    djn *pScan ,#TIME ; after 19 self-scans
    jmp *pScan ,}tWipe ; switch to dat-wiping
    dat 0 ,0
    tSpl spl #1 ,{1
    dat 0 ,0
    dat 0 ,0
    tEnd dat 0 ,0

    for MAXLENGTH-CURLINE-2-1
    dat 0 ,0
    rof

    ;tDecoy equ (tWipe+7007)
    ;tStart mov <tDecoy+0 ,{tDecoy+2 ; make a quick-decoy
    ; mov <tDecoy+3 ,{tDecoy+5 ; to foil one-shots
    ; mov <tDecoy+6 ,{tDecoy+8 ; and the occasional q-scan
    ; djn.f tScan+1 ,<tDecoy+10

    B equ (tWipe+7007)

    tStart mov.i <B ,{B+3
    I for 1
    mov.i <B+1+(3*I) ,{(B+(3*I)+3)
    rof
    djn.f tScan+1 ,<B+5+(3*1)

    end tStart

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Stephen Gunnell@21:1/5 to Stephen Gunnell on Sat Oct 29 19:08:20 2022
    On Sunday, 30 October 2022 at 7:16:48 am UTC+8, Stephen Gunnell wrote:

    ;name He Scans Alone qhoz
    tScan sub.x #0-STEP ,tPtr ; increment and look

    It is probably worth noting that I kept STEP as a fairly small value as did Paul in his original code ( the optimiser selected 10/8 in qhoz and the original was 12/9 ). During the development of Aoshi I also noted that using different steps for the A and
    B fields was very effective. I also note that a mod 10 step seems to also be a value that my Hazy Lazy scanners gravitate towards. If you set STEP to be a large mod 10 value the A field may efficiently scan the core and when it found something it would
    pass the value to the B field which would scan forwards in small increments.

    Steve

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From John Metcalf@21:1/5 to Stephen Gunnell on Sun Nov 6 18:35:20 2022
    On Sat, 29 Oct 2022 19:08:20 -0700, Stephen Gunnell wrote:

    On Sunday, 30 October 2022 at 7:16:48 am UTC+8, Stephen Gunnell wrote:
    It is probably worth noting that I kept STEP as a fairly small value as
    did Paul in his original code ( the optimiser selected 10/8 in qhoz and
    the original was 12/9 ). During the development of Aoshi I also noted
    that using different steps for the A and B fields was very effective. I
    also note that a mod 10 step seems to also be a value that my Hazy Lazy scanners gravitate towards. If you set STEP to be a large mod 10 value
    the A field may efficiently scan the core and when it found something it would pass the value to the B field which would scan forwards in small increments.

    Steve

    Congrats, impressive score on 94nop :-) I'd be interested to compare
    which warriors your code selected with those in my hand-picked benchmark.

    There are a few tweaks in some other HSA-style scanners which might be
    worth trying:

    Seven-Eleven (.8c, steps 7/11), DAT 0, {0 for the dat-wipe

    Razor (.66c, steps 7/9), perfect gate, SNE.I scanner

    Willow (.66c, steps 13/9), DAT >0, 0 for the dat-wipe

    Ash (.8c, steps 17/11), use SNE.X and alternates between SUB.F and SUB.X

    Trouble (.8c, steps 11/16), start scan at 3rd instruction after self-scan

    Here's the code for Trouble:

    ;redcode-94nop
    ;name Trouble
    ;author John Metcalf
    ;strategy scanner
    ;assert CORESIZE==8000

    x equ 3100
    y equ 600
    stepx equ 11
    stepy equ 16
    count equ 16

    ptr dat y, x+y-1

    for stepx-8
    dat 0, 0
    rof

    dat >0, 0
    sb spl #0, {0

    wipe mov @bp, <ptr
    mov >ptr, >ptr
    jmn.f wipe, >ptr

    scan sub.x inc, ptr
    sne.x @ptr, *ptr
    inc sub.x #-stepy, ptr
    jmn.f hit, @ptr
    jmz.f scan, *ptr

    mov.x @scan, @scan
    hit slt.b @scan, #last+2-ptr
    djn.f wipe, ptr
    djn inc, #count
    bp djn.f inc, #sb

    last end scan+1

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Stephen Gunnell@21:1/5 to John Metcalf on Sun Nov 6 14:50:40 2022
    On Monday, 7 November 2022 at 2:35:22 am UTC+8, John Metcalf wrote:

    Congrats, impressive score on 94nop :-) I'd be interested to compare
    which warriors your code selected with those in my hand-picked benchmark.

    The current list is:
    Starting WarriorZ ... lore2.red - 0.963639104067866
    Starting Her IMPertinent Majesty ... yellownewspaper.red - 0.968492684024264 Starting Visigoth ... sharky.red - 0.97160387599864
    Starting biSHO ... k310.red - 0.971778549072255
    Starting ShadowWeaver ... borgir.red - 0.969684633279653
    Starting dreadful cross ... lastjudgement.red - 0.977390396931096
    Starting AshII ... hsaopt-qhoh.red - 0.979644545493337
    Starting Resinoid ... lordimprings2.red - 0.963729691567763
    Starting King Cobra ... twilight.red - 0.935340724763019
    Starting Tolyputes ... blowrag2.red - 0.966682391867335
    Starting Despair ... borgir.red - 0.970876224369984
    Starting Pocket Rocket ... zplusplus.red - 0.929591489465872
    Starting Electra ... lordimprings2.red - 0.974757202343255
    Starting Monster Human Grunt ... elvenking2.red - 0.975380831472071
    Starting incubus ... hullabaloo3.red - 0.969622420270168

    The number at the end is the correlation coefficient between the selected proxy warrior's scores and the scores of the real warrior.

    Here is the corresponding list for 94x:
    Starting DriedFrogPillsAX9 ... dfpopt-qreo.red - 0.988565232237808
    Starting PuddleglumX ... puddleglum-xjcl.red - 0.99928754183971
    Starting paperX ... grimoireA-ugbc.red - 0.991370859383051
    Starting DriedFrogPillsAX133 ... dfpopt-nxac.red - 0.995470490449125
    Starting Morlock ... deadletter-xlwm.red - 0.931948655588622
    Starting PendulumX ... memories07x.red - 0.932070614564449
    Starting BlackMoods ... dflyoptB-prrt.red - 0.887106456201055
    Starting Slayer ... mavinopt1-dtat.red - 0.93102766140238
    Starting DespairX ... deadletter-tiru.red - 0.98989167508046

    I left PuddleglumX floating as a sanity check and the program has correctly found the correct version. For the two copies of Dried Frog Pills the originals were deleted and I let the program select stand ins.
    As you can see with Black Moods I have no good approximations in my repertoire.

    I'm happy to publish the code if people are interested but it is in Perl.

    Steve

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Stephen Gunnell@21:1/5 to Stephen Gunnell on Tue Nov 8 19:55:24 2022
    On Monday, 7 November 2022 at 6:50:41 am UTC+8, Stephen Gunnell wrote:

    Starting Resinoid ... lordimprings2.red - 0.963729691567763
    Starting Electra ... lordimprings2.red - 0.974757202343255

    I got annoyed at multiple unknown warriors mapping to the same proxy so I added code to allow unknown warriors to differentiate from each other.
    94nop is now:
    Starting Monster Human Grunt ... elvenking2.red - 0.975380831472071
    Starting biSHO ... k310.red - 0.971778549072255
    Starting incubus ... spiritual.red - 0.970529104184943
    Starting Despair ... borgir.red - 0.970876224369984
    Starting WarriorZ ... lore2.red - 0.951270800865769
    Starting Resinoid ... unheardof.red - 0.954543201503372
    Starting dreadful cross ... lastjudgement.red - 0.977390396931095
    Starting Pocket Rocket ... zplusplus.red - 0.929591489465872
    Starting King Cobra ... twilight.red - 0.935340724763019
    Starting Her IMPertinent Majesty ... yellownewspaper.red - 0.968492684024264 Starting Electra ... lordimprings2.red - 0.980289980569215
    Starting Visigoth ... sharky.red - 0.97160387599864
    Starting Tolyputes ... blowrag2.red - 0.966682391867335
    Starting ShadowWeaver ... borgir.red - 0.969684633279653

    Mind you, I'm struggling to keep up with all John's submissions and it isn't helping that messages from the hill are randomly dissapearing.

    Steve

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From John Metcalf@21:1/5 to Stephen Gunnell on Mon Nov 21 14:02:13 2022
    On Sun, 06 Nov 2022 14:50:40 -0800, Stephen Gunnell wrote:

    On Monday, 7 November 2022 at 2:35:22 am UTC+8, John Metcalf wrote:
    I'm happy to publish the code if people are interested but it is in
    Perl.

    I'd be interested :-) It appears to be doing a good job of matching the warriors to a proxy.

    John

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)